在 2022/3/25 9:09, Jakub Kicinski 写道:
On Thu, 24 Mar 2022 23:49:14 +0800 Jian Shen wrote:
Introduce a set of bitmap operation helpers for netdev features, then we can use them to replace the logical operation with them. As the nic driversare not supposed to modify netdev_features directly, it also introduces wrappers helpers to this.
The implementation of these helpers are based on the old prototype of netdev_features_t is still u64. I will rewrite them on the last patch, when the prototype changes.
Signed-off-by: Jian Shen <shenjian15@huawei.com> --- include/linux/netdevice.h | 597 ++++++++++++++++++++++++++++++++++++++ Please move these helpers to a new header file which won't be included by netdevice.h and include it at users appropriately. ok.
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index 7307b9553bcf..0af4b26896d6 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -2295,6 +2295,603 @@ struct net_device { }; #define to_net_dev(d) container_of(d, struct net_device, dev)
+static inline void netdev_features_zero(netdev_features_t *dst) +{ + *dst = 0; +} + +static inline void netdev_features_fill(netdev_features_t *dst) +{ + *dst = ~0ULL; +} + +static inline bool netdev_features_empty(const netdev_features_t src) Don't pass by value something larger than 64 bit. That's a problem. There are many functions in kernel already using netdev_features_t as input parameter or return value. For example:
/netdev_features_t (*ndo_fix_features)(struct net_device *dev, netdev_features_t features);/ If change the prototype of netdev_features_t from u64 to unsigned long *, we have to move the return value as a output parameter. Is there any other way to fix this ?
+{ + return src == 0; +} + +/* helpers for netdev features '==' operation */ +static inline bool netdev_features_equal(const netdev_features_t src1, + const netdev_features_t src2) +{ + return src1 == src2; +} +/* helpers for netdev features '&=' operation */ +static inline void +netdev_features_direct_and(netdev_features_t *dst, + const netdev_features_t features) +{ + *dst = netdev_features_and(*dst, features); +} + +static inline void +netdev_active_features_direct_and(struct net_device *ndev, s/direct_and/mask/ ? ok
+ const netdev_features_t features) +{ + ndev->active_features = netdev_active_features_and(ndev, features); +} + +/* helpers for netdev features '|' operation */ +static inline netdev_features_t +netdev_features_or(const netdev_features_t a, const netdev_features_t b) +{ + return a | b; +} +/* helpers for netdev features '|=' operation */ +static inline void +netdev_features_direct_or(netdev_features_t *dst, s/direct_or/set/ ?
ok
+ const netdev_features_t features) +{ + *dst = netdev_features_or(*dst, features); +} +/* helpers for netdev features '^' operation */ +static inline netdev_features_t +netdev_features_xor(const netdev_features_t a, const netdev_features_t b) +{ + return a ^ b; +} +/* helpers for netdev features '^=' operation */ +static inline void +netdev_active_features_direct_xor(struct net_device *ndev, s/direct_xor/toggle/ ? ok
+/* helpers for netdev features '& ~' operation */ +static inline netdev_features_t +netdev_features_andnot(const netdev_features_t a, const netdev_features_t b) +{ + return a & ~b; +} +static inline void +netdev_features_direct_andnot(netdev_features_t *dst, s/andnot/clear/ ? ok + const netdev_features_t features) +{ + *dst = netdev_features_andnot(*dst, features); +} +/* helpers for netdev features 'set bit' operation */ +static inline void netdev_features_set_bit(int nr, netdev_features_t *src) s/features_set_bit/feature_add/ ? It's ok to change set_bit to add, should I remove 's' for single bit ? +{ + *src |= __NETIF_F_BIT(nr); +} +/* helpers for netdev features 'set bit array' operation */ +static inline void netdev_features_set_array(const int *array, int array_size, + netdev_features_t *dst) +{ + int i; + + for (i = 0; i < array_size; i++) + netdev_features_set_bit(array[i], dst); +} + +#define netdev_active_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->active_features) + +#define netdev_hw_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->hw_features) + +#define netdev_wanted_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->wanted_features) + +#define netdev_vlan_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->vlan_features) + +#define netdev_hw_enc_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->hw_enc_features) + +#define netdev_mpls_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->mpls_features) + +#define netdev_gso_partial_features_set_array(ndev, array, array_size) \ + netdev_features_set_array(array, array_size, &ndev->gso_partial_features) + +/* helpers for netdev features 'clear bit' operation */ +static inline void netdev_features_clear_bit(int nr, netdev_features_t *src) All the mentions of '_bit' are unnecessary IMHO. . thanks, this make the name of helper more succinct.
summarize : netdev_features_direct_and -> netdev_features_mask netdev_features_direct_or -> netdev_features_set netdev_features_direct_xor -> netdev_features_toggle netdev_features_direct_andnot -> netdev_features_clear netdev_features_set_bit -> netdev_feature_add netdev_features_clear_bit -> netdev_feature_del netdev_features_test_bit -> netdev_feature_test